A-MEM: Self-Evolving Memory for Coding Agents

Introduction

A-MEM is an innovative self-evolving memory system specifically designed for coding agents. Unlike traditional vector stores, A-MEM automatically organizes knowledge into a dynamic Zettelkasten-style graph, where memories don't just get stored, they evolve and connect over time. This system aims to provide agents with a more intelligent and interconnected knowledge base, improving their ability to recall and utilize information effectively.

Installation

Getting started with A-MEM is straightforward. You can install it via pip and integrate it with MCP-compatible agents like Claude Code.

First, install the a-mem package:

pip install a-mem

Then, add it to your Claude Code environment:

claude mcp add a-mem -s user -- a-mem-mcp \
  -e LLM_BACKEND=openai \
  -e LLM_MODEL=gpt-4o-mini \
  -e OPENAI_API_KEY=sk-...

A session-start hook will automatically install, prompting Claude to use the memory system.

Examples

A-MEM exposes several tools for coding agents and can also be used directly via its Python API.

Agent Tool Usage (Conceptual):

# Store a memory (returns task_id immediately)
add_memory_note(content="Auth uses JWT in httpOnly cookies, validated by AuthMiddleware")

# Search later
search_memories(query="authentication flow", k=5)

# Deep search with connections
search_memories_agentic(query="security", k=5)

Python API Usage:

from agentic_memory.memory_system import AgenticMemorySystem

memory = AgenticMemorySystem(
    llm_backend="openai",
    llm_model="gpt-4o-mini"
)

# Add (auto-generates keywords, tags, context)
memory_id = memory.add_note("FastAPI app uses dependency injection for DB sessions")

# Search
results = memory.search("database patterns", k=5)

# Read full details
note = memory.read(memory_id)
print(note.keywords, note.tags, note.links)

Why Use A-MEM?

A-MEM offers several compelling features that enhance the capabilities of coding agents:

• Self-Evolving Memory: Memories are not static. When new knowledge is added, A-MEM automatically finds related memories, strengthens connections, updates context, and evolves tags, creating a dynamic knowledge graph.
• Semantic + Structural Search: It combines vector similarity with graph traversal, allowing agents to find memories by meaning and then explore their interconnected relationships for deeper insights.
• Peek and Drill: This feature enables efficient token usage by first capturing relevant memories via lightweight metadata (ID, context, keywords, tags) using breadth-first search. Agents can then drill depth-first into specific memories for full content, maximizing recall while minimizing token costs.

Links

• GitHub Repository: https://github.com/DiaaAj/a-mem-mcp
• PyPI: https://pypi.org/project/a-mem/